About
AI governance divides into two distinct problems. The first is epistemic: what happens at the inference layer, where models produce outputs that are more or less truthful, more or less accountable to evidence. The second is structural: the legal, political, and institutional forces that shape what labs build, what they are permitted to do, and what governance they will accept. Most attention goes to the second. The first is undersupplied.
That split became visible through direct experience building against it. The work started as a personal irritation with AI systems making confident, sloppy arguments — false equivalences, epistemic mush dressed up as balance. The obvious move was to stop complaining and understand why. That led to architecture, architecture led to governance, and governance turned out to be a shape problem more than a configuration problem. The runtime contract is what emerged: a governed inference layer that enforces evidence discipline, uncertainty boundaries, and structural void handling at the inference layer itself. EID — Expectation-Induced Determinism — is a phenomenon observed in that process, not a feature designed into it.
That work addresses what engineering can reach at the epistemic layer. But it operates inside a system it does not control. Structural governance — law, regulation, liability exposure, political pressure — determines whether epistemic intervention ever gets traction at scale. Labs respond to those forces whether or not the forces understand what they are governing. The entanglement is not incidental. It only becomes visible at full scope, and full scope is the only place from which you can expect to move anything.
That is why both halves are in scope here. Covering the epistemic layer without covering the structural forces that constrain it would be an omission on a site about epistemic accountability.
2ndlaw is the work of Eric Soldan, an independent researcher focused on epistemic governance for AI systems. The research, the runtime contract, the perspectives, and the lexicon are his. Direct technical conversation: [email protected].
If you are building systems and want to know how 2ndlaw actually fits:
These pages assume you care about concrete behavior and access rather than theory for its own sake.
If you are more interested in the underlying ideas:
These are attempts to describe why the runtime contract looks the way it does, and what kinds of epistemic failure modes it is meant to address. There are also additional writings (under Perspectives).
If you want to engage directly:
There are no forms or funnels. If you have something concrete to talk about, email is the interface.